Generating Neural Networks to Detect Alzheimer’s
Business + ManagementConferencesData Science for GoodHealthcareHealthcareneural networkdsWest 2018posted by Elizabeth Wallace, ODSC August 5, 2019 Elizabeth Wallace, ODSC
This post discusses how neural networks can help detect Alzheimer’s.
AI is showing so much promise in the medical field. It’s an excellent example of how AI combines with human intelligence to create a “super brain” capable of predicting disease, uncovering patterns, and testing solutions for persistent problems. Precision Medicine is a medical nonprofit using this super combination of AI and human expertise to find solutions to one of the most devastating diseases we know, Alzheimer’s. Ayin Vala, co-founder and someone who lost his own grandfather to this illness, outlines Precision Medicine’s work.
[Related Article: Machine Learning and Compression Systems in Communications and Healthcare]
We Need Data Scientists and Healthcare Experts
There are no treatments to prevent the disease and no cure. It’s a growing problem. Unlike other persistent diseases such as HIV or Heart Disease, in which people are dying less now than in previous generations, Alzheimer’s deaths are going up. It’s now the sixth leading cause of death in the US and the most expensive single disease to treat.
While it’s not all bad news, there are still no significant advancements in Alzheimer’s overall. We are making headway for early prevention and treatment, and early detection is currently our biggest hope for slowing the disease. Precision Medicine is massaging the data into research that could help make new headway.
To do research with AI in healthcare, you need three things. You have to cultivate and curate your data despite privacy laws and siloed data inputs. A lot of our data isn’t available or hasn’t been processed yet.
You also need novel data science methods to overcome this data difficulty. Healthcare data is notoriously unstructured and private. Plus, you’re addressing shortcomings already in the field of healthcare like slow research turnover or lack of expertise.
The process needs institutional, clinical, and research partners. People connect their healthcare expertise with someone’s knowledge in data science and take advantage of funding from an institution with the resources both need. Research must be collaborative.
Medical Precision’s goal was to get as much data as it could. It was in a unique position as a nonprofit to get the kind of data they needed, but data was in different silos and would need to be processed. The company was able to procure around four million patient records, including Brain MRI scans from over 2000 patients.
Medical Precision wants to perfect early detection using structured data and MRI Images and possibly implement this ability into an app.
The Process for Alzheimer’s Clinical Research
Medical Precision began with research curated from hundreds of papers related to Alzheimer’s, including certain medications and conditions that are highly correlated with the disease’s onset. For example, we know that diabetes is highly correlated with Alzheimer’s diagnoses.
Structured data comes in, and the company puts it through machine learning or neural networks to find patterns and accurate predictions for the disease. Early detection using this structured data extracted features like age, gender, etc. and used those in addition to other indications (medications and other conditions). The variables move through the algorithms to find those predictions, but the company also wants to uncover revelations in contributing factors as well. The accuracy is pretty good – .8 to .85 AUC – but the company is pursuing greater efficiency with each iteration of their research.
The Tools Medical Precision Uses
The company uses Google Cloud, including Big Query, Data Studio, and Kubernetes Engine, among others. They envisioned the dashboard similar to what pilots use to fly planes safely and efficiently. However, it’s difficult to put this kind of dashboard in front of physicians. They’re busy. They may not want to work with another tool. Medical Precision was able to pivot quickly to an Alzheimer’s prevention app that puts information directly into patients’ hands. It gives the patient key factors to begin a conversation with a doctor about their individual risks for Alzheimer’s.
Medical Precision isn’t trying to replace doctors. Instead, they’re trying to facilitate early detection by giving patients their own app for understanding the risks. Medical doctors can then take those risks and build plans for prevention.
The Future of Alzheimer’s Prevention
Doctors have about a 95% accuracy rating for predicting and diagnosing Alzheimer’s in patients, but that depends on patients following the steps for testing in the first place. Medical Precision is hoping to increase this accuracy for more patients overall, helping to get more people onto an early treatment plan before the most devastating effects of the disease are underway.
[Related Article: ODSC East 2019: Major Applications of AI in Healthcare]
In healthcare, accuracy isn’t always the most important thing right off. Vala acknowledges that it’s better to be less sensitive with predictions and screen a larger number of patients than risk sending patients home with incorrect screenings. That will bring accuracy down a bit, but as machines get better with human help, we can still see both accuracies while keeping the umbrella as wide as we can.
Hopefully, with this kind of work, we’ll see fewer and fewer families devastated by Alzheimer’s and other diseases. Early detection is currently the best hope we have of mitigating some of Alzheimer’s worst effects, and machines could boost our ability to predict who’s at risk.